无人机群多目标协同主动感知的自组织映射方法

针对主动感知问题多为单机器人系统的主动视觉问题，本文提出了基于自组织映射特征网络的异构机器人主动感知框架，为无人机团队规划出遍历所有目标所需时间最短的平滑路径。首先把多目标主动感知场景建模为带邻域的多旅行商问题，然后使用自组织映射网络为无人机团队规划出旅行时间最短的闭环轨迹，最后利用三阶贝塞尔曲线对轨迹做平滑处理。仿真结果和对比实验表明，本文的方法在多目标主动感知的应用中有着较好的效果。

Self-organizing feature map method for multi-target active perception of unmanned aerial vehicle systems

It is known that the robots fitted with cameras, also sometimes called as unmanned aerial vehicles, are used to monitor inaccessible areas. The data sent by robots are analyzed through various artificial neural networks. Active vision is particularly important to cope with problems like occlusions, limited field of view, and limited resolution of the camera. In view of the active vision difficulty faced in active perception system, this paper proposes an active perception framework of heterogeneous robots based on the self-organizing feature map, a type of multi-objective active learning algorithm, and plans the shortest smooth path for the drone team to traverse all the targets. First, the multi-target active perception scene is modeled as multiple traveling salesmen problem with neighborhood. After that, the self-organizing mapping network is used to find the shortest closed-loop track for travel time, and then the track is smoothed with the third-order Bézier curve, a parametric curve. The simulation results and comparative experiments show that proposed method has better effects in the application of multi-objective active perception.